Academic Speaker Series - Kris Ferreira
The CMU INFORMS Student Chapter, jointly with Tepper OM Group, hosted Prof. Kris Ferreira, Assistant Professor at Harvard Business School on Nov 8 2019. In her research, Prof. Ferreira draws inspiration from real industry problems. She presented her work on Learning to Rank an Assortment of Products which was an outcome of her collaboration with Wayfair. Sunanda Parthasarathy, Associate Director of Algorithms/Data Science at Wayfair Inc., and Shreyas Sekar, Post Doctoral Fellow at Harvard University are co-authors in this work.
With the rise of online retailers like Amazon, we have witnessed an exponential rise of interest around assortment optimization, which involves choosing the set of products to display to a customer such that the expected revenue of the seller is maximized. Prof. Ferreira however looks at this problem from a different angle. She wishes to maximize customer engagement in Wayfair’s website (which is a departure from the revenue maximizing scheme in assortment optimization). According to the model, if the customer does not like any of the products within her attention window, she leaves the site without browsing any additional products. Otherwise, the customer is hooked and continues browsing. The goal therefore is to maximize the number of hooked customers.
The authors first analyze the complete information case in which the seller (Wayfair) knows the joint distribution of click probabilities of customers for various products and the attention window. The problem is shown to be NP hard. They propose a greedy algorithm which achieves an approximation ratio of 1/2. It is also straightforward to see that no algorithm can do better than the ratio 1-1/e. They also show for a special case in which the click probabilities of a customer are independent of her attention window, the greedy algorithm does achieve the 1-1/e ratio.
Then the authors focus on the more realistic case in which the seller has to learn the joint distribution of click probabilities and the attention window while displaying products. They propose two learn-then-earn algorithms. The first one, which they call the Simple Learning Algorithm is based on the Multi Armed Bandit approach which obtains good theoretical guarantees with a slow convergence rate. The second algorithm is the novel Threshold Acceptance Algorithm which is based on the idea of exhaustively trying all products at a given rank in order to identify the best one, the algorithm simply selects a product that is good enough. The latter algorithm achieves similar guarantees with (significantly) faster convergence.
Other than the very informative talk, the Ph.D. students benefitted a lot from the conversations during their meetings and at lunch with Prof. Ferreira. In her conversations, she highlighted the advantages and challenges of working with industry partners, and shared some of her experiences about her collaborations. We thank Prof. Ferreira for visiting us and sharing her work.
